<h2>Must know</h2>
<ul>
<li><strong>Multi cluster GKE Inference Gateway enters preview</strong>: Google Cloud has launched a multi-cluster Inference Gateway to scale AI workloads across regions, addressing GPU and TPU resource silos. The architecture utilizes Kubernetes Custom Resources like InferencePool and GCPBackendPolicy to enable model-aware load balancing and capacity bursting. According to Google Cloud, this setup provides global low-latency serving and automated disaster recovery for production-grade inference. <a href="https://app.daily.dev/posts/79zWldGPG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes reliability is now a machine speed problem</strong>: At scale, Kubernetes failures emerge from cascading interactions between autoscalers and GitOps controllers that unfold faster than human response times. Cloud Native Now reports that platform teams are increasingly deploying AI-driven agentic investigation layers to correlate telemetry and state changes in real time. This shift moves SREs into high-level decision-making roles while autonomous agents handle initial triage at machine speed. <a href="https://app.daily.dev/posts/MJPxr5Icj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uber bypasses etcd limits with MySQL backed ML mesh</strong>: Uber’s Michelangelo platform has evolved into a global mesh handling 30 million predictions per second by utilizing over 100 custom CRDs. To bypass standard etcd scaling constraints, the team implemented a MySQL-backed storage abstraction and a federated batch scheduling layer. The New Stack notes that future iterations will focus on autonomous self-healing via AI agents for debugging and zero-toil framework upgrades. <a href="https://app.daily.dev/posts/rfWF0zX0x" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>ClickHouse Full text Search reaches general availability</strong>: Native inverted index support now enables 7–10x faster cold query performance for log search and observability workloads. <a href="https://app.daily.dev/posts/wNpt02khc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubermatic 2.30 integrates Kubernetes MCP Server</strong>: The new release enables LLM-assisted cluster management and replaces NGINX Ingress with native Gateway API support. <a href="https://app.daily.dev/posts/uCWkpGrR9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare launches Custom Regions for data sovereignty</strong>: Enterprise customers can now define precise geographical boundaries for TLS termination and Layer 7 processing to meet strict compliance needs. <a href="https://app.daily.dev/posts/BQn3C61sU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA DSX Air simulates AI factories in the cloud</strong>: This SaaS platform allows teams to validate full-stack AI infrastructure, including networking and storage, before physical hardware deployment. <a href="https://app.daily.dev/posts/mGEKXvn3S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Argo CD powered kubara framework simplifies cluster bootstrapping</strong>: This open-source Go CLI tool can set up a production-ready multi-cluster Kubernetes platform with default security policies in under 30 minutes. <a href="https://app.daily.dev/posts/4wTpimtvB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft provides guidance for Ray on AKS</strong>: The new reference architecture uses Azure Arc for GPU quota aggregation and BlobFuse2 for POSIX-compliant shared ML storage. <a href="https://app.daily.dev/posts/NHWWMUVCd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vite 8.0 replaces Rollup with Rust based Rolldown</strong>: The architectural shift promises 10-30x faster builds and includes integrated DevTools for deep build analysis. <a href="https://app.daily.dev/posts/FK7SZspfC" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 100x engineer gap is the new management myth</strong>: A recent opinion piece claims AI has widened the gap between great and mediocre engineers from 2x to 100x. This is a classic case of ‘vibe-based’ metrics that ignore the reality that 80% of engineering is still human coordination and architectural judgment. It’s a convenient narrative for managers looking to justify aggressive headcount cuts while pretending a chatbot can replace a senior architect. <a href="https://app.daily.dev/posts/o9h36HY82" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Orbital datacenters are peak infrastructure insanity</strong>: Nvidia and partners are hyping ‘space-based computing’ for geospatial AI, but Gartner analysts correctly label this as peak insanity. Prohibitive launch costs and the minor detail of cooling high-performance GPUs in a vacuum make this more of a PR stunt than a viable infrastructure strategy. We haven’t even solved terrestrial power grid bottlenecks yet, but sure, let’s put a Blackwell rack in orbit. <a href="https://app.daily.dev/posts/5QtI91Gu3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are not a substitute for a real SRE</strong>: Vendors are pushing ‘AgenticOps’ as a way for AI to autonomously reason and act on network issues. While these tools are great for summarizing logs, giving them write access to production without a massive safety harness is just asking for an automated outage. As seen with internal incidents at Amazon, an over-privileged agent is just a very fast way to delete your entire production environment. <a href="https://app.daily.dev/feed-by-ids?id=4z4AhzJNR&amp;id=Nu2d4YyjL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
